Using recent results from numerical relativity simulations of black hole mergers, we revisit previous studies of cosmological black hole spin evolution. We show that mergers are very unlikely to yield large spins, unless alignment of the spins of the merging holes with the orbital angular momentum is very efficient. We analyze the spin evolution in three specific scenarios: (1) spin evolves only through mergers, (2) spin evolves through mergers and prolonged accretion episodes, and (3) spin evolves through mergers and short-lived (chaotic) accretion episodes. We study how different diagnostics can distinguish between these evolutionary scenarios, assessing the discriminating power of gravitational-wave measurements and X-ray spectroscopy. Gravitational radiation can produce three different types of spin measurements, yielding, respectively, the spins of the two black holes in a binary inspiral prior to merger, the spin of the merger remnant (as encoded in the ring-down waves), and the spin of ''single'' black holes during the extreme mass-ratio inspiral (EMRI ) of compact objects. The latter spin population is also accessible to iron-line measurements. We compute and compare the spin distributions relevant for these different observations. If iron-line measurements and gravitational-wave observations of EMRIs only yield dimensionless spins j ¼ J /M 2 > 0:9, then prolonged accretion should be responsible for spin-up, and chaotic accretion scenarios would be very unlikely. If only a fraction of the whole population of low-redshift black holes spins rapidly, spin-alignment during binary mergers (rather than prolonged accretion) could be responsible for spin-ups.